To address the issues that the traditional algorithms have achieved low accuracy, and research on solving track association tasks from the perspective of interactive analysis is still lacking, a interactive analysis approach for ship target track association and fusion under multi-source detection is proposed. Firstly, the nearest neighbor distance track association algorithm is improved for interrupted track association and multi-source track association. Secondly, the visual analysis advantages of “human in the loop”are fully utilized, and the weight of the sensor in the trajectory fusion process is calculated by evaluating its stability. Finally, the interrupted trajectory stitching method based on interpolation fitting and the multi-source track association method based on weighted average are used for trajectory fusion. The effectiveness of the proposed algorithm in enhancing the correct rate of interrupted track association and reducing the error rate of multi-source track association is demonstrated throught the results of experiments. The designed interactive analysis system can verify the effectiveness of the fusion algorithm. Through interactive analysis and improved correlation fusion algorithm, the task of interrupted track and multi-source track association can be completed more accurately.
对于中断航迹关联,算法流程如下:1)箱线图检测航迹的距离矩阵 D,矩阵内的元素初始化为正无穷(N代表传感器内有效航迹总数);2)生成同维度关联矩阵 A,各位置初始值设为0;3)遍历待关联传感器下的所有航迹,当前航迹索引为i,设置为旧航迹,再遍历待关联传感器下的所有航迹,当前航迹索引为j,设置为新航迹。若新航迹的起始时间大于旧航迹的终止时间,则计算航迹i和航迹j之间所有参考属性之间的距离。将 D 中位置处的值更新为,得到传感器内所有有效航迹间的距离矩阵 D :
从航迹对之间的中断时间和旧航迹末端信息与新航迹首端信息的关系出发对原始基于最近邻距离的中断航迹关联算法在中断时间关系条件判断和关联距离计算上进行改进,其中中断时间基于参数调节过程设置,关联距离基于旧航迹的末端信息和新航迹的首端信息计算所得,包括航迹1终点和航迹2起点之间的空间欧式距离、航向角度距离、航速数值距离。改进算法的执行流程如下:1)创建维距离矩阵 D,所有元素初始值为正无穷;2)生成同维度关联矩阵 A,各位置初始值设为0;3)循环访问传感器内所有航迹,选定当前航迹编号i后再次遍历全部航迹,若航迹j满足起始时刻与航迹i终止时刻的时间差超过预设阈值,则计算二者关联距离,将距离矩阵 D 位置处的值更新为;4)定位距离矩阵 D 的最小值坐标,将关联矩阵 A [i][j]置为1,并将 D 矩阵第i行与第j列元素赋值为无穷大;5)循环执行步骤4)直至 D 中所有元素处理完毕,此时关联矩阵 A 即当前传感器中断航迹的成对关联结果。算法距离矩阵更新流程如图6所示。
流程如下:1)输入序列和,初始化维累计距离矩阵 D,、分别为两序列长度;2)迭代访问序列的每个点i,计算其与首点的累计距离,赋值 D 中处为;3)同理迭代访问序列的每个点,计算其与首点的累计距离,赋值 D 中处为;4)双重循环遍历所有点对,计算第个点与第个点的累计距离,更新 D 中处为。
式中,为序列第个点与序列第个点之间的距离。改进算法的实现流程如下:1)创建维距离矩阵 D,所有元素初始化为正无穷(、分别表示传感器1、2的航迹数量);2)生成维关联矩阵 A,全矩阵初始化为0;3)双重循环遍历传感器1的航迹(索引为)与传感器2的航迹(索引为)。4)若航迹与的采样点数量比例(较大者/较小者)低于预设阈值,则调用动态时间规整算法;5)计算航迹对的关联距离,并更新距离矩阵 D 中处为;重复步骤4)直到 D 中所有元素均为,此时 A 矩阵即为多源航迹的成对关联结果。距离矩阵更新流程如图7所示。
XIONGW, XUP L, CUIY Q, et al. Track segment association via track graph representation learning[J]. IET Radar, Sonar & Navigation, 2021,15(11):1458-1471.
[5]
CAOY P, CAOJ W, ZHOUZ G. Track segment association method based on bidirectional track prediction and fuzzy analysis[J]. Aerospace, 2022,9(5):274.
[6]
WANGJ, ZENGY J, WEIS M, et al. Multi-sensor track-to-track association and spatial registration algorithm under incomplete measurements[J]. IEEE Transactions on Signal Processing, 2021,69:3337-3350.
WILLEMSN, VAN HAGEW R, DE VRIESG, et al. An integrated approach for visual analysis of a multisource moving objects knowledge base[J]. International Journal of Geographical Information Science, 2010,24(10):1543-1558.
[12]
ANJ F, QIAOT T, YANGX, et al. Design of a visual analysis platform for sea route based on AIS data[C]∥Proceedings of the 2019 2nd International Conference on Artificial Intelligence and Big Data. Piscataway, USA: IEEE, 2019:102-106.